Effect of streamflow forecast uncertainty on real-time reservoir operation
Research highlights
► A dynamic uncertainty evolution model is used to characterize streamflow forecast. ► Modeled forecast uncertainty is incorporated to dynamic reservoir operation. ► Ensemble and probabilistic forecasts have high potential for improving decisions.
Introduction
Advances in weather forecasting, hydrologic modeling, and hydro-climatic teleconnection relationships have significantly improved streamflow forecast precision and lead-time [3], [22], [24], [28] and provide great opportunities to improve the efficiency of water resources system operations [23], [25], [29], [39]. In recent years, forecast products, particularly long-term streamflow forecasts (with a lead-time longer than 15 days), have been applied to reservoir operation and water resources management (e.g. [23], [25], [29], [39]).
In addition to forecast precision and lead-time, operation strategies also influence the efficiency of utilizing streamflow forecasts for real-time reservoir operation [4], [20], [39]. As a common practice, reservoir operation curves, which set a target storage level for each operation period around a year, are adopted as guidelines for real-time reservoir operation as well as for operation planning [18], [34]. Since operation curves are determined by historical streamflow records [20], [34], they reflect suitable reservoir operation decisions under various historical scenarios rather than real-time streamflow conditions. Thus, even a perfect streamflow forecast cannot improve reservoir operation efficiency when operation curves are used [39]. In many recent studies, reservoir operation curves have been replaced by real-time reservoir optimization and simulation models, which are supposed to provide more flexible and efficient approaches utilizing various streamflow forecast products [8].
One important issue with implementing streamflow forecasts in real-time reservoir operation models is dealing with the uncertainty involved in streamflow forecast products [8], [9], [26]. Although forecast uncertainty analysis has been one research focus in hydrology (e.g. [17], [31], [32]), there are comparatively less studies on the effect of forecast uncertainty on real-time reservoir operations [9], [27], [33]. Deterministic or probabilistic streamflow forecast products are usually treated as ad hoc inputs for deterministic or stochastic reservoir operation models. That is to say, a deterministic forecast or a stochastic forecast represented by a number of scenarios is pre-designed for a specific reservoir operation problem for screening test, and no non-generalizable structure of the forecast error is endogenously involved in the operation analysis. Correspondingly, many previous studies on forecast and reservoir operation in the literature adopt a two-component approach, one provides (“recommends”) a forecast scenario [3], [22], [24], [28] as input to the other component [23], [25], [29], [39] that dealing with forecast application. In general, such an approach suggests that forecast can always improve reservoir operation efficiency especially under extreme conditions [21].
This study aims at analyzing the effect of forecast uncertainty on real-time reservoir operations. As different forecast products, e.g., deterministic and probabilistic streamflow forecasts, can exert different effects on real-time reservoir operation decisions in optimization and simulation models, this study will explicitly simulate the uncertainty in each of the streamflow forecasts examined and assess its effect on real-time reservoir operation decisions. Since the tool for such a purpose does not exist in the hydrologic literature, the Martingale Model of Forecasting Evolution (MMFE) [11], [12] used in supply chain management is introduced to quantify real-time streamflow forecast uncertainty and generate deterministic and probabilistic forecast products. Simulations based on standard operation policy (SOP), dynamic programming (DP), and stochastic dynamic programming (SDP) [16], [18] are adopted to determine release decisions for a hypothetical reservoir using synthetic streamflow forecasting products.
The rest of the paper is organized as follows. Section 2 provides some background information on streamflow forecasting and forecast uncertainty and introduces the Martingale Model of Forecasting Evolution (MMFE). Section 3 describes the MMFE-based forecast uncertainty analysis in real-time reservoir operation. Section 4 introduces the numerical experiments designed in this study. Section 5 analyzes the results and Section 6 contains the conclusions.
Section snippets
Background
In hydrology, there are various indices reflecting the magnitude of streamflow forecast uncertainty (e.g., [24], [32]). However, few models illustrate the forecast uncertainty evolution process. This paper adopts MMFE from supply chain management [11], [12] to quantify the evolution of the uncertainty of real-time streamflow forecasts as time progresses.
MMFE-based streamflow forecast uncertainty analysis
To use MMFE to model the uncertainty of streamflow forecasts, it is necessary to justify its assumptions, i.e. unbiasedness, non inter-period correlation, stationarity, and Gaussian distribution. Real-time streamflow forecasts are based on hydrologic model inputs, such as precipitation, temperature, and soil moisture. These inputs are updated at the beginning of each period with new weather forecasts and hydrologic observations (e.g., streamflow, soil moisture) to improve the preceding
Numerical experiments
A hypothetical reservoir system with N operation periods (i.e., studying horizon of the operation problem) is used in this study. In reservoir operation, the forecast lead time H is assumed to be the same as the length of remaining operation periods (i.e., the lead time H is N periods at the beginning, N − 1 periods when decision moves to next period, and so on). SOP, DP, and SDP models are then used to generate reservoir operation decisions with various synthetic streamflow forecast products.
Result analysis
The effect of streamflow forecast uncertainty on real-time reservoir operation is analyzed with reservoir operation models DP, SDP, and SOP. In the context of forecast uncertainty analysis, the effect of streamflow variability and reservoir capacity are also assessed under a pre-specified forecast uncertainty level, as shown in Table 1.
Conclusions
Streamflow forecast uncertainty plays an important role in reservoir operation, but the effects of forecast uncertainty on reservoir operation have yet to be thoroughly addressed in a unifying framework. Rather than treating the forecast products as ad hoc inputs to reservoir operation models, this study provides a method to characterize the forecast uncertainty evolution and explicitly assess the effect of streamflow forecast uncertainty on real-time reservoir operation. The Martingale Model
Acknowledgements
This research was partially supported by the China Overseas Scholarship Foundation, the National Natural Science Foundation of China (Project No. 50928901) and the US National Aeronautics and Space Administration (NASA) grant (Project No. NNX08AL94G). The authors are grateful for the numerous helpful suggestions from Dr. Mohamad Hejazi on the early versions of this paper.
References (41)
- et al.
How to simulate normal data sets with the desired correlation structure
Chemom Intell Lab Syst
(2010) - et al.
Assessment of Folsom Lake response to historical and potential future climate scenarios: 1. Forecasting
J Hydrol
(2001) - et al.
Evaluation of stochastic reservoir operation optimization models
Adv Water Resour
(2009) - et al.
Ensemble flood forecasting: a review
J Hydrol
(2009) - et al.
Reservoir optimization using sampling SDP with ensemble streamflow prediction (ESP) forecasts
J Hydrol
(2001) - et al.
Towards the characterization of streamflow simulation uncertainty through multimodel ensembles
J Hydrol
(2004) - et al.
The role of hydrologic information in reservoir operation – learning from historical releases
Adv Water Resour
(2008) - et al.
Quantifying hydrological modeling errors through a mixture of normal distributions
Journal of Hydrology
(2007) - et al.
Assessment of Folsom Lake response to historical and potential future climate scenarios. 2. Reservoir management
J Hydrol
(2001) An experimental study of the effectiveness of rolling schedules in production planning
Decision Sci
(1977)
Short-term, single, multi-purpose reservoir operation – importance of loss functions and forecast errors
Water Resour Res
Optimal hedging and carryover storage value
J Water Res Plan Manage – ASCE
Potential benefits of seasonal inflow prediction uncertainty for reservoir release decisions
J Appl Meteorol Climatol
A tactical planning-model for a job shop
Oper Res
Modeling the evolution of demand forecasts with application to safety stock analysis in production distribution systems
IIE Trans
Approximate solutions of a dynamic forecast-inventory model
M&SOM
Comparison of stochastic and deterministic dynamic-programming for reservoir operating rule generation
Water Resour Bull
Sampling stochastic dynamic-programming applied to reservoir operation
Water Resour Res
Bayesian theory of probabilistic forecasting via deterministic hydrologic model
Water Resour Res
Optimal operation of multireservoir systems: state-of-the-art review
J Water Res Plan Manage – ASCE
Cited by (156)
Estimation of water storage capacity of Chinese reservoirs by statistical and machine learning models
2024, Journal of HydrologyModel predictive control and rainfall Uncertainties: Performance and risk analysis for drainage systems
2024, Journal of HydrologyExploiting multiple hydrologic forecasts to inform real-time reservoir operation for drought mitigation
2023, Journal of Hydrology